Fine-tune models with LoRA/QLoRA and implement secure model distribution.
| # | Type | Title | Platform | Duration |
|---|
| 5.1 | Video | When to Fine-Tune vs RAG | Concept | 10 min |
| 5.2 | Video | Databricks Fine-Tuning | Databricks | 10 min |
| 5.3 | Lab | Fine-Tune in Databricks | Databricks | 40 min |
| 5.4 | Video | LoRA/QLoRA from Scratch | Sovereign | 10 min |
| 5.5 | Lab | Fine-Tune with entrenar | Sovereign | 45 min |
| 5.6 | Video | Model Encryption and Signing | Sovereign | 10 min |
| 5.7 | Lab | Secure Model Pipeline with pacha | Sovereign | 35 min |
| 5.8 | Video | EU AI Act and Governance | Concept | 8 min |
| 5.9 | Quiz | Fine-Tuning and Security | — | 15 min |
entrenar for LoRA/QLoRA trainingpacha for ChaCha20-Poly1305 encryption
- Freeze base model weights
- Add trainable low-rank matrices
- Scaling factor:
alpha / r
- Target modules: q_proj, v_proj, k_proj
- Quantized base model (4-bit)
- Double quantization for memory efficiency
- Paged optimizers for large batches
| Aspect | Fine-Tuning | RAG |
|---|
| Knowledge | Baked into weights | Retrieved at runtime |
| Updates | Requires retraining | Update index only |
| Cost | Higher compute | Lower compute |
| Use case | Style/behavior change | Knowledge access |